Alice cinematic input consumption: bounded diagnostic

Run: scratch/runs/20261118-alice-cinematic-input-consumption-cause/. Predecessor: prepared player control.

The startup Escape reached the native input handler and entered its queue, but the original made no USER dequeue attempt during the five-second intent. The same queue edge remained until cancellation. There was no renderer input poll, native dequeue candidate, or native return candidate; the armed native USER observation sequence remained zero. The bounded trace dropped no rows. This is a polling gap, rather than evidence of a PM_NOREMOVE, target, or receipt validator defect. The independent physical release was acknowledged promptly, so synchronous control-event-loop starvation is not supported by this trace. UNKNOWN closed that scope; there were no subsequent endpoint requests or replay.

A fresh launch selected passive startup before its timer began. Without Escape, the authentic original reached its main menu after approximately 244 seconds. Reviewed Enter selected New Game, then reviewed Enter selected Easy. Both were consumed by the real owning USER PeekMessage return (hwnd 65543, thread 1, PM_NOREMOVE), in approximately 850 and 980 milliseconds, with independent UP acknowledgements. Thus the unchanged native path can deliver ordinary original menu input once the guest polls. No presentation-count threshold was used as readiness.

Pandemonium loading subsequently progressed to the Cheshire Cat loading screen, then remained there through the selected gameplay cutoff. Read-only observations showed changing execution addresses around 0x468f01–0x46910d and an unchanged rendering bridge (474 presents). During the Quit reserve the scene eventually advanced to autonomous floating alphabet blocks. Reviewed ordinary Escape then consumed normally and the captured consequence was an Alice title frame. It did not establish a Quit confirmation. The retained registers and stack identify a concrete next investigation: guest work during level loading, including the byte/bit decoding loop around those addresses. No causal player W, jump, attack, or ordinary Quit was demonstrated. Harness cancellation or deadline termination is not ordinary Quit. The original lifetime, independent watchdog, and 180-second Quit reserve were fixed before launch; they were not extended.

No Alice runtime patch is justified by these observations, and no Alice runtime or controller code was changed. In particular, increasing an input timeout would not repair a guest that is not polling. A browser-versus-native original comparison and DirectInput/GetAsyncKeyState timing remain unproved; the general consumer trace exhausted its detailed budget before the relevant interval. Its later USER samples must not be presented as continuous DirectInput tracing.

Evidence limits: the first Escape's scene annotation incorrectly anticipated Rogue; the actual reviewed file showed the EA Games cinematic. Viewing and dispatch had been batched, so personal review did not precede that dispatch. original-diagnostic/review-correction.json preserves this error. It does not invalidate the physical queue trace, but that dispatch is not reviewed gameplay evidence. Diagnostic trace identity fields based on the hidden first console also must not be treated as the actual game target; readiness and input snapshots identify the real game window. Two passive observation subprocesses ended with SIGTERM; retained partial observations are not complete observation intervals.

The run-local check-cause.cjs regression checks unchanged queued Escape, absence of dequeue observations, UNKNOWN closure, and the same original's exact native menu/Easy receipts plus independent releases. The three focused Alice helper tests passed. This is evidence validation, not a synthetic guest-poll reproducer or an assertion that player control works.

The required full build initially detected a stale generated lib/gl-gps-declines.generated.js after an incoming GLGPS change. Regenerating that file reconciles reason 10's explanatory text with its existing source and JSON mirror. This separate build repair changes diagnostic text, not Alice's polling behavior. Its generator check and the final locked build are retained.

Recovery uses Git, the predecessor's retained exact module, and the existing authentic media package. Module SHA-256: 46c9b8e4af91a8770d454ff529e21d6a17553c31ad57aea30ca4a8aa6bf3e311. Base runtime Git: 2b8ba1b675a6fc0ce286d48a6261dc04acef62dd. Committed override revision: 5732320bb39c4b7f7327f51e30c6ed1d07e26074; the exact three override hashes and frozen graph inventory are retained in initial-git.json and each scope's runtime-inputs.json. Both observed source graphs and all 16 media pins passed verify-inputs.cjs. No foreign WIP or media copy was adopted. recovery-reference.json, publication.json, and result.json record recovery paths, final actual waits, and the final outcome. No box was created.